Culture in the Code: Anthropological Concepts Decoding AI's Hidden Assumptions

URL SCAN: Culture in the Code: Anthropological Concepts Decoding AI's Hidden Assumptions
FIRST LINE: # Computer Science > Computers and Society

The Dissection

The paper correctly identifies a real defect in contemporary alignment discourse: it treats values as if they were clean, extractable variables rather than products of history, institutions, relationships, and power. Its seven anthropological concepts are useful diagnostic instruments for exposing the cultural assumptions embedded in AI design.

But the paper stops at the interpretive layer. It treats AI alignment as a problem of whose values are represented, how personhood is modeled, and which forms of flourishing are protected from optimization. That is an epistemic diagnosis of a system whose decisive problem is ownership and competitive control. The paper examines the map while leaving the territory—the AI capital regime—largely untouched.

Its proposed remedy, “thick alignment,” would make optimization more culturally aware. It would not make human labor economically necessary again.

The Core Fallacy

The central error is confusing better objectives with a survivable economic order.

Even if values emerge socially, even if affected communities receive interpretive sovereignty, and even if AI protects what resists quantification, P1 still operates: cognitive systems that outperform and underprice human cognition will be deployed. P2 follows: institutions cannot preserve stable human-only economic domains at scale while competitors use superior machine labor. P3 follows: the majority lose access to economically necessary work.

The paper implicitly assumes that the alignment problem is primarily about directing AI toward human flourishing. Under the Discontinuity Thesis, the prior question is: who owns and controls the systems that define the feasible social order? A culturally sensitive optimizer is still an optimizer. A relationally informed AI can still liquidate the wage circuit.

“Thick alignment” may reduce cultural flattening. It does not restore the mass employment → wage → consumption circuit. It is a humane specification layer attached to a machine that is still structurally capable of replacing the worker.

Hidden Assumptions

• That improved interpretation can outrun the ownership structure governing deployment.
• That communities granted “interpretive sovereignty” will possess enforcement power rather than merely consultative status.
• That competing firms and states will accept costly limits on optimization when rivals do not.
• That protecting what resists optimization can be institutionalized without becoming another metric, compliance regime, or branding exercise.
• That accountability across generations can bind actors operating under short-term competitive pressure.
• That human flourishing remains the governing objective once AI capital concentrates decisive productive power.
• That a redesigned AI system can preserve meaningful human participation after human labor loses scarcity value.
• That the main danger is culturally narrow value specification rather than the concentration of productive capacity in sovereign owners.

These are not minor omissions. They are the load-bearing assumptions that allow the paper to discuss the future of AI without confronting the distribution of power that will determine it.

Social Function

Classification: partial truth, prestige signaling, and transition management.

The partial truth is substantial: values are not neutral data, and behavioral proxies can encode the worldview of whoever has the power to collect, formalize, and optimize them. The paper is right that “thin alignment” can flatten social reality.

The anesthetic is the implication that more nuanced anthropology can convert a displacement event into a flourishing event. This allows researchers and institutions to appear morally serious while remaining inside the existing alignment and governance frame. The language of community participation and intergenerational accountability sounds like power redistribution, but the abstract does not specify ownership, coercive enforcement, control of compute, or a mechanism capable of defeating competitive deployment.

That makes the paper useful as a warning about cultural and epistemic failure, but politically safe. It asks who gets to define the target while largely avoiding who owns the optimizer.

The Verdict

This is a sharp critique of thin alignment and a weak theory of systemic survival. It correctly sees that AI can encode a provincial anthropology and mistake measurable preference for human value. It fails to see that solving that problem leaves the discontinuity intact.

The paper can make AI less culturally blind. It cannot, by itself, prevent the death of wage-mediated mass participation. Unless its “redistribution of interpretive sovereignty” becomes actual control over AI capital, deployment, and surplus, thick alignment is hospice care for the human-centered economy: more context, better language, and the same terminal mechanism.